Windows Dictation Privacy and Online Speech Recognition

On Windows 11, voice typing uses online speech recognition ¹. With Online speech recognition turned on, your voice data is sent to Microsoft to be turned into text ². Voice Access is different: Microsoft describes it as working without an internet connection ³, once its language files are installed during setup ⁴. Whether Microsoft may keep your voice to improve its speech recognition is a third choice, the voice clip setting. It is optional, and work and school accounts cannot use it at all ⁵. If you handle health records, Microsoft's published HIPAA list does not name Windows voice typing or Voice Access, so a privacy switch is not enough.

Voice typing is processed in Microsoft's cloud, Voice Access on your PC

Voice typing and Voice Access use different speech engines, and the difference decides where your audio goes. Microsoft says voice typing uses online speech recognition, which is powered by Azure Speech services ⁶. It also needs an internet connection, a working microphone and the cursor in a text box ⁷.

In practice, the audio you speak is what travels. Microsoft names the purpose of that trip: voice data goes to Microsoft "only to provide the service and create text transcriptions" ². Nothing in Microsoft's voice typing description says the audio is kept after the transcript comes back, unless you contribute it as a voice clip. That is why the clip setting, covered below, is the setting to check first.

Voice Access does its work on the PC. Its setup downloads the language files for the language you choose, which are "used by on-device speech recognition" ⁴. The internet is needed for that first download and for finishing setup ⁸. Once the files are installed, Voice Access works without an internet connection ³.

Fluid dictation edits your words before they land in the text box, and that edit step stays on the PC. If you want the original wording back, Microsoft's page says you can say "Revert" or "Undo that" ( Microsoft's fluid dictation page). Those commands only undo the correction. They do not change where the audio went.

The table shows one gap that matters. Microsoft's pages on Voice Access name the on-device engine, but Microsoft's speech privacy article never mentions Voice Access at all. Our answer for Voice Access rests on Microsoft's Voice Access pages, not on the privacy article. The full answer to that question is on our page about whether Voice Access sends audio to Microsoft.

Online speech recognition is the setting that sends your voice to Microsoft

The switch that controls the cloud path is Online speech recognition. Microsoft lists the steps in its speech privacy article for Windows 11:

1. Select Start, then Settings.
2. Select Privacy & security, then Speech.
3. Set Online speech recognition to On or Off.

Microsoft describes two kinds of recognition. Device-based recognition processes your voice locally, and no voice data goes to Microsoft ⁹. Online recognition uses Microsoft's cloud for more accurate results. Microsoft says voice data is sent only to provide the service and create transcriptions, and that it does not store, sample or listen to recordings without your permission ². That last clause is why the voice clip setting matters. It is the route Microsoft describes for keeping your audio.

That is also the trade-off. Microsoft presents the cloud path as giving more accurate recognition, so turning the switch off means accepting what device-based recognition gives you. Keep that in mind when you judge the choice for a particular job.

Contributing voice clips is a separate, optional choice

A voice clip is a recording of what you say, which Microsoft may sample and listen to in order to improve its speech recognition. Contributing one is a different switch from online speech recognition. Microsoft says you can use voice typing without contributing voice clips, and that contributing them is optional ¹¹. To change the setting:

1. Press Windows logo key + H to open voice typing.
2. Select Settings.
3. Follow the links to start or stop contributing voice clips ¹².

The same steps let you stop again. Turning contribution off does not change the voice typing feature itself, which keeps working either way ¹¹.

Once you contribute, the rules change in four ways. Each one comes from Microsoft's support article on voice data.

• Retention. Microsoft keeps contributed clips for up to two years ¹³. A clip that has been sampled may be kept longer, so that Microsoft can keep training its models ¹⁴.
• Review. Microsoft employees and vendors working for Microsoft can review contributed clips ¹⁵.
• De-identification. Microsoft says its software removes the link between a clip and your account or device identifiers ¹⁶. That protects who the recording is linked to. It does not change what the recording contains.
• Background sound. Microsoft warns that other sounds audible during a recording may be recorded too ¹⁷. A clip made at a desk with a colleague talking nearby can capture that colleague.

The contribution rules have a dated starting point. Microsoft says it stopped logging voice data for product improvement on October 30, 2020 ¹⁸. The support article is undated in the copy we read, so the two-year figure may have changed. Check Microsoft's current page before you decide.

Work and school accounts cannot contribute voice clips

If you are signed in with a work or school account, Microsoft says sharing voice data is not available, but you can still use voice typing ⁵. The clip setting is therefore not a choice an employee can make on a managed account. Voice typing on that account still uses online speech recognition, so voice data still goes to Microsoft ².

Microsoft's page does not say what happens when someone uses a personal account on a work PC, so ask your IT team about that case as well. We did not verify whether an IT administrator can turn off online speech recognition by policy. Microsoft's pages we read do not say. Ask your IT team before you assume the switch is available. Apps also need microphone access, which you can review under Privacy & security, then Microphone, as Microsoft explains in its microphone permissions article.

Turning off online speech recognition stops voice typing but not on-device features

Turning off Online speech recognition leaves only device-based features, and Microsoft gives Narrator as the example ¹⁹. Voice typing depends on online recognition ¹, so it stops.

Microsoft does not say whether Voice Access is among the features that remain. We did not test it. If you need Voice Access to work with online recognition off, check it yourself:

1. Turn off Online speech recognition in Privacy & security, then Speech.
2. Turn on Voice Access and say a few commands, such as opening an app or clicking a button by name.
3. Dictate a sentence into a text box, then check your network activity with a tool you trust.

If Voice Access keeps working and the network activity stays quiet, you have evidence for your own machine. It is not a guarantee from Microsoft, and it may change with a Windows update. Other on-device features follow the same pattern. Microsoft says live captions process audio and create captions on the device, and that audio, voice data and captions are not shared with Microsoft.

Fluid dictation's language model runs on the device

Fluid dictation corrects grammar, punctuation and filler words as you speak, and Microsoft says it is "powered by on-device small language models (SLMs)" ¹⁰. It is a Copilot+ PC feature for English locales, according to Microsoft's fluid dictation page. It is switched off automatically in secure fields such as passwords or PINs ²⁰.

The page describes the correction step as on-device. It does not say whether the speech recognition underneath is on-device or online. Fluid dictation belongs to voice typing, which uses online speech recognition ¹, so treat the audio path as online until Microsoft says otherwise. Only the correction step is confirmed as local. For more on the feature, see fluid dictation on Copilot+ PCs.

Regulated work such as healthcare needs more than a privacy setting

HIPAA rules apply to what you do with protected health information, and a privacy switch does not change them. A covered entity may let a business associate handle electronic protected health information only after it obtains satisfactory assurances that the business associate will safeguard it ²¹. HHS guidance says a cloud service provider that holds electronic protected health information for a covered entity must enter into a HIPAA-compliant business associate agreement with it ²². We could not open HHS's current page, so this cites an archived copy.

Microsoft says its business associate agreement is available by default to covered entities and business associates ²³, and that it covers in-scope services ²⁴. The list we read on October 16, 2026 names Azure, Dynamics 365 and Windows 365, but it does not name Windows voice typing or Voice Access ( Microsoft's HIPAA page). Its Office 365 table does list a Speech microservice ²⁵. The table does not say which feature that microservice serves, so do not read it as covering dictation in Windows.

A signed agreement is also not the end of the job. Microsoft says the agreement helps support your compliance, but using its services does not achieve HIPAA compliance on its own ²⁶. The Security Rule requires an accurate and thorough risk analysis of the electronic protected health information you hold ²⁷.

Before you dictate protected health information on a Windows PC, put these questions to Microsoft and get the answers in writing:

1. Is Windows voice typing or Voice Access covered by a business associate agreement for the account type you use?
2. Does any audio from dictation leave the PC, and if so, is any of it kept after the transcript is returned?
3. Can an administrator turn off online speech recognition and voice clip contribution for your accounts?
4. Is the speech recognition under fluid dictation online or on the device?

Our answer to whether Windows dictation is HIPAA compliant: on October 16, 2026, we could not find any Microsoft statement that covers Windows voice typing or Voice Access under a business associate agreement. A covered entity should not assume compliance. Get the answer in writing from Microsoft and from your compliance team before you dictate protected health information. This is general information, not legal advice. The question page on Windows dictation and HIPAA covers the same ground, and the guide on HIPAA compliant dictation covers the wider choice.

The most private option is dictation that never leaves the PC

Microsoft's developer speech service handles audio differently from the Windows features. For real-time speech to text in Azure, the audio is "processed only on the Azure's server memory," and no data is stored at rest ²⁸. That page describes the Azure service that developers call. It does not describe what happens to audio you dictate in Windows, so we do not assume the two behave the same.

Compare the paths side by side. Each row is the data path Microsoft describes on its pages, and the last row is for developers, not for dictation in Windows.

The voice typing row combines two statements from Microsoft. It is our reading of them together, not a single statement from Microsoft. For ordinary contributed clips, two years is the period Microsoft states, but a sampled clip may be kept longer.

If the audio has to stay on the PC, the choice narrows. Voice typing covers 46 languages and regional variants ²⁹, but it runs online. Voice Access covers 15 languages and dialects ³⁰ and runs on the PC. Microsoft's Voice Access FAQ gives a shorter list of languages, so its pages disagree. This guide uses the setup page. Microsoft groups the two lists differently, so read the difference as a rough guide.

To keep your audio off Microsoft's servers on Windows:

1. Turn off Online speech recognition in Privacy & security, then Speech.
2. Leave voice clip contribution off in voice typing Settings.
3. Use Voice Access for dictation, and test it with the switch off before you rely on it.
4. If the work is regulated, get Microsoft's and your compliance team's answer in writing first.

Yap, which we make, is an on-device dictation app for Mac and Windows. Compare it with the built-in tools on the same privacy questions. For the Mac side of the same question, read how Apple handles dictation privacy. For how long voice data is kept across products, see the guide on voice data retention. Other guides in our Windows dictation overview cover offline speech to text on Windows, voice typing, Voice Access and the languages each dictation feature supports.